Skip to content

BACKLOG.md

Latest commit

 

History

History
169 lines (149 loc) · 9.58 KB

BACKLOG.md

File metadata and controls

169 lines (149 loc) · 9.58 KB
Raw

TiaCAD Backlog

Consolidated open action items scattered across the doc set, gathered during a doc-coherence pass on 2026-07-11. This is the single place to look for "what's left to do" that isn't the headline roadmap item — see ROADMAP.md for the Constraint Solver (Q4 2026) and CGA v5.0 (2027+) strategic track.

How this file works: one line per item, themed, with a pointer to the doc/code that has the full context. Items move to CHANGELOG.md when shipped (delete from here), not the other way around. Re-verify an item before starting it — several of these (marked below) haven't been checked against current code since the date given.

Validation / testing infrastructure

  • CI validation as a required gate. Add schema-conformance + schema-truth checks; make expect: contract checking a required CI gate, not just a local tiacad check --contract command. — VALIDATION_STRENGTHENING.md §5
  • One source of truth for test health. — shipped 2026-07-11: scripts/generate_test_status.py parses the non-visual suite's junit-xml + coverage.xml (plus a --collect-only count of the visual suite) into a committed TEST_STATUS.json. tests.yml's python-3.11 leg generates it every run and, on push to main, commits it back ([skip ci], contents: write job permission) — check this file for current pass/fail/skip/coverage counts instead of hand-writing them into markdown. — VALIDATION_STRENGTHENING.md §5
  • Differential testing blocked. ~90% of geometry code bypasses the GeometryBackend abstraction and calls CadQuery directly, so the fast MockBackend can't stand in for most tests and true kernel-vs-kernel differential testing is currently impossible. Routing new operation code through the backend is a prerequisite. Same root cause as the CadQuery coupling item below. — VALIDATION_STRENGTHENING.md §5, KNOWN_LIMITATIONS.md #3
  • Pinned CI lockfile. Exact dependency versions, regenerated on review — decouples "what the library supports" from "what CI reproducibly tests." Explicitly deferred when the dependency-posture pass shipped 2026-07-10. — VALIDATION_STRENGTHENING.md §5
  • Trust-gallery sign-off (§4.7) and golden STEP set (§4.9) — both explicitly out of scope for the T0-T5 ladder, still open. — VALIDATION_STRENGTHENING.md §6
  • Curved-geometry watertight false negative. Mesh-export tessellation artifact on spheres/fillets makes the watertight check unreliable for curved surfaces — worth fixing before more Tier-3-style manifold-health gates lean on it. — KNOWN_LIMITATIONS.md #8
  • validation: schema block: implement or delete. — resolved 2026-07-11: the schema no longer has a validation: key (top-level properties are …parts, operations, finishing, export, variants, expect); the shipped expect: contract block superseded it. (The schema_version drift half was likewise done — 0 files still say "2.0".)

Model-validation UX (docs/developer/MODEL_VALIDATION.md, items 2-7)

  • Reconcile item 2 ("model-local contracts") against the now-shipped expect: contract engine — likely partially/fully superseded, needs a doc pass before carrying forward as open.
  • tiacad verify CLI command — evaluate model-local contracts, emit JSON + console summary.
  • Reference-based measurements — distances/angles/alignment between named spatial references, exposed as a CLI/testing utility.
  • Stepwise summaries attached to operations in build_trace.json.
  • Annotated trust renders — point at measured failures directly on the rendered image.
  • Negative trust scenarios — intentionally-bad models that must fail visual/trust-render validation specifically (distinct from the Tier 5 parse/build negative corpus, which already covers parse-time failures).

API / language surface

  • Deprecation warnings for old syntax. — shipped 2026-07-11: all four legacy patterns (cone radius_bottom/radius_top, linear-pattern scalar spacing+direction, translate: {offset:} wrapper, list-form export:) now raise a runtime DeprecationWarning with a migration pointer and a backward-compat mapping. 15 tests in test_parser/test_deprecation_warnings.py. — docs/developer/API_DEPRECATION_STRATEGY.md
  • GitHub import: no branch override. — shipped 2026-07-11: trailing @branch suffix now selects the branch (default main, slashes allowed), branch-namespaced in the cache. — KNOWN_LIMITATIONS.md #4
  • Limited export formats. No DXF/IGES/G-code/SVG — community-driven, add on demand. — KNOWN_LIMITATIONS.md #2

Architecture debt

All items below re-verified against current code on 2026-07-11 (see git log / session weightless-universe-0711). None turned out fully stale — every claim has at least partial current-code support — but four had their framing narrowed. Originals from docs/archive/OPEN_ISSUES_AND_ARCHITECTURE_DEBT_2026-04-18.md (now archived — see "Docs hygiene" below).

Still fully true:

  • Backend global state. Selection still falls back to process-global state (get_default_backend/set_default_backend/reset_default_backend in geometry/__init__.py); production call sites in parts_builder.py and backend_utils.py still depend on the ambient default rather than an injected backend. (TiaCADParser entry points do accept explicit injection now — the global fallback path is what remains.)
  • Part coupling hub. part.py (237 lines) is confirmed the #1 fan-in file in the repo (21 importers within tiacad_core/, 59 repo-wide) — small file, largest blast radius. Purely a coupling problem, not size/complexity.
  • cli.py monolith. Still the #1 quality hotspot in the repo (1,201 lines, 57 functions, 59/100); create_parser() alone is 122 lines, and TiaCADParser/DimensionError are re-imported locally in 6+ command functions instead of once at module scope.
  • tiacad_core.visual vs tiacad_core.visualization. Both packages still exist with overlapping purpose (trust/debug rendering vs. general rendering); no canonical-boundary decision documented. Three non-code sibling dirs (visual_references/, visual_output/, visual_diffs/) widen the naming confusion though they aren't part of the package-boundary problem itself.

True but narrower than originally framed:

  • OperationsBuilder dispatch. execute_operation() still routes via a 12-branch if/elif on op_type (not a registry), but the operation logic has already been delegated to per-type builder classes — this is now a routing-mechanism issue, not a god-object one.

  • Parse orchestration. parse_pipeline.py remains a confirmed hotspot (parse_tiacad_dict is 110 lines / depth 4 doing 6 distinct jobs). watcher.py is moderately complex (start/_rebuild/_resolve_watch_paths all 40–70 lines) but is not among the repo's top hotspots — lower priority than parse_pipeline.py; don't treat both as equally heavy.

  • Validation heuristic vs semantic. Assembly-level rules (validation/rules/*.py) are still bbox-heuristic-heavy (5 of 9 rule files lean on bbox/bounds helpers), but a real semantic layer now ships and is used — the expect: contract engine (testing/contracts.py, spatial-relation + manifold-overlap checks) appears in 53+ example/test files. The claim is now accurate only for the rule-based validator, not the codebase as a whole.

  • parameter_resolver.py. High-coupling module (3rd-highest fan-in in the repo) but not a complexity hotspot by Reveal's thresholds — only _check_cycles() (complexity 10, depth 6, mixing DFS + regex + error formatting) stands out. Track as coupling/consolidation risk, not complexity.

  • Schema/version contract drift — fixed 2026-07-10, verified 2026-07-11 (0 files declare schema_version: "2.0").

Docs hygiene

  • CadQuery coupling. STL/STEP export and several advanced operations still require CadQuery-compatible geometry rather than going through GeometryBackend — same root cause as the differential-testing blocker above; worth tracking as one item, not three separate mentions across docs. — KNOWN_LIMITATIONS.md #3
  • docs/architecture/README.md / ARCHITECTURE_NEXT_STEPS.md / OPEN_ISSUES_AND_ARCHITECTURE_DEBT_2026-04-18.md overlap. — resolved 2026-07-11: OPEN_ISSUES_AND_ARCHITECTURE_DEBT_2026-04-18.md and its same-date near-duplicate ARCHITECTURE_REVIEW_2026-04-18.md (both debt registers covering identical findings) archived to docs/archive/ — their content is superseded by this file's own re-verified "Architecture debt" section above. ARCHITECTURE_NEXT_STEPS.md kept as the canonical forward-looking plan (genuinely distinct content: recommendations/roadmap, not a debt register). docs/architecture/README.md, docs/README.md, and docs/DOCUMENTATION_MAP.md updated to point here instead.
  • Two archive-index files — done 2026-07-11: deleted the orphaned docs/archive/README.md duplicate; docs/archive/ARCHIVE_SUMMARY.md is now the single canonical archive index (already the one docs/DOCUMENTATION_MAP.md pointed to).

Items resolved during the 2026-07-11 doc-coherence pass have already been removed from their source docs and are not repeated here — see CHANGELOG.md's Unreleased section and git log for that session's diff if you want the full list of what got fixed (DAG doc status contradiction, stale example-count/version numbers, three conflicting constraint-solver quarter labels, docs/DOCUMENTATION_MAP.md regeneration, several archived docs).